Customer value analysis: A two-stage data mining approach

Customer relationship management (CRM) has long been regarded as an important problem to understand and measure the true value of customers. In particular, churn management is one major task of CRM to retain valuable customers since to retain valuable customers is much more important than to obtain new (but may not be valuable) customers. However, this leads to a research problem of effectively identifying valuable customers for churn management. As data mining techniques have been widely used in recent literature to discover useful information and/or knowledge from a huge amount of data, this paper considers a two-stage data mining approach to analyze the value of customers. Specifically, this paper takes an automobile parts company as an example, and the first stage uses association rules and decision trees to select representative variables from the chosen dataset respectively. Next, the second stage uses decision trees to develop the customer value analysis model based on the output produced by the first stage. The experimental results by comparing decision trees alone, association rules + decision trees, and decision trees + decision trees show that combining two-stage of decision trees not only provides the highest rate of prediction accuracy (81.6%), but also reduces the original 26 variables to 5 representative values. On the other hand, combining association rules and decision trees provides the lowest Type I error, which means that this model has the lowest error rate of recognizing valuable customers into non-valuable customers.